Skip to content

Stan Weekly Roundup, 11 August 2017

This week, more Stan!

  • Charles Margossian is rock star of the week, finishing off the algebraic solver math library fixture and getting all plumbed through Stan and documented. Now you can solve nonlinear sets of equations and get derivatives with the implicit function theorem all as part of defining your log density. There is a chapter in the revised manual

  • The entire Stan Development Team, spearheaded by Ben Goodrich needing fixes for RStan, is about to roll out Stan 2.17 along with the interfaces. Look for that to begin trickling out next week. This will fix some further install and error message reporting issues as well as include the algebraic solver. We are also working on moving things toward Stan 3 behind the scenes. We won’t just keep incrementing 2.x forever!

  • Ben Goodrich fixed the inlining declarations in C++ to allow multiple Stan models to be linked or built in a single translation unit. This will be part of the 2.17 release.

  • Sean Talts is still working on build issues after Travis changed some config and compilers changed everywhere disrupting our continuous integration.

  • Sean is working with Michael Betancourt on the Cook-Gelman-Rubin diagnostic and have gotten to the bottom of the quantization errors (usingly on 1000 posterior draws and splitting into too many bins).

  • Imad Ali is looking ahead to spatio-temporal models as he wraps up the spatial models in RStanArm.

  • Yuling Yao and Aki Vehtari finished the stacking paper (like model averaging), adding references, etc.

  • Yuling has also made some updates to the loo package that are coming soon.

  • Andrew Gelman wrote papers (with me and Michael) about R-hat and effective sample size and a paper on how priors can only be understood in the context of the likelihood.
  • Jonah Gabry, Ben and Imad have been working on the paper for priors based on $R^2$

  • Andrew, Breck Baldwin and I are trying to put together some educational proposals for NSF to teach intro stats, and we’re writing more grants now that it’s grant season. All kinds of interesting things at NSF and NIH with spatial modeling.

  • Jonah Gabry continues the ggplot-ification of the new Gelman and Hill book.

  • Ben Goodrich has been working on multivariate effective sample size calculations.

  • Breck Baldwin has been working with Michael on courses before Stan (and elsewhere).

  • Jonah Gabry has been patching all the packagedown references for our doc and generally cleaning things up with cross-references and organization.

  • Mitzi Morris finished adding a data qualifier to function arguments to allow propagation of data-only-ness as required by our ODE functions and now the algebraic solver.

  • Dootika Vats was visiting Michael; she’s been working on multivariate effective sample sizes, which considers how bad things can be with linear combinations of parameters; scaling is quadratic, so there are practical issues. She has previously worked on efficient and robust effective sample size calculations which look promising for inclusion in Stan.

  • Bob Carpenter has been working on exposing properties of variables in a Stan program in the C++ model at runtime and statically.


  1. Richard McElreath says:

    Wow re the algebraic solver.

    Excited also about the stacking paper, since stacking is basically what I encouraged in my book (blending posterior predictives using CV weights), but without any solid justification. It’s been done in biology for years, just as a sort of folk tradition. If I understand right, the new proposed weights are better, since they use the score distribution more effectively.

  2. Sebastian says:

    +1 for that algebraic solver ! Huge step for Stan in applications I work on.

    … and also great to see the advancements on fishing out static stuff which is so important for MPI.

  3. Rahul says:

    >>>Bob Carpenter has been working on exposing properties of variables in a Stan program in the C++ model at runtime and statically.<<<

    What's the larger goal here? Just curious.

Leave a Reply